MarkMarine

This assumes flock is just doing ALPR and it’s obvious from police statements and stolen flock camera data dumps they are well beyond that now, onto facial recognition.

The govt has been steadily eroding privacy and our rights (carnivore reading email, bulk collection of phone metadata, all internet traffic with x-keyscore, stingrays, etc etc etc) and using terrorism as the boogeyman when often it’s just wrong think that will put you in the sights. None of this needed warrants, all of it illegal by a plain reading of the 4th amendment. If a constitutional amendment won’t protect you, a state law is basically toilet paper.

I would argue they need 1 product owner for the app surface so I can’t see their org structure so clearly. Those jagged edges and users hitting blocks depending on what section of the app they were in are design failures, I know how a computer works and when using the Claude app it wasn’t always clear to me what the right combo of things to press to get my desired outcomes.

For me it’s still jagged, but chatgpt never had the weird “cowork” abstraction with the Linux machine.

I don’t talk to non-technical users that use the chatGPT app for work so I’m not sure if that’s been a positive change for them, but I would imagine if that was put in place for Claude it would be better for them. Chat is instant answers, different model, faster; work is more detailed, thinks more, has access to all the tools. This is what I see the users doing, but there is a third mode they use (deep research, exhaustively find me the right answer to a crucial question) that is underserved right now.

All of this is moot to me as a dev, I use terminal, herdr, pi… so I don’t see the jagged edges myself. But regular users do

Claude has a jagged UX if you support non-devs. The surfaces before today in the desktop app were: Chat, cowork, Claude code.

Inside Claude code you’ve got cloud environments and local.

Let’s say as a normal desktop user you wanted to just automate something normal in your work, going to a website (maybe some internal app at your company) getting some data. Depending on what surface you used this will either not work, not work well (cowork), or work quite well but possibly be blocked because of bot controls (local Claude code) or again not work (cloud Claude code)

It was nuts. You can clearly tell these were different teams and products mashed into the same app. Super confusing for non technical users. Hopefully it’s a little better today

there are a lot of gas generators powering people's camps.

This depends on your users and doc corpus. On finance docs and with users that use jargon and acronyms heavily, BM25 can fail on trivial queries... you end up encoding a masters in economics into the query re-writing logic.

Double hard if you're dealing with private market finance customers with their own ideas on what "common" terms mean. I tried to replace the embedding/ingestion pipeline multiple times and nothing I tried was better over a large amount of documents. Performance sucked, the agent was re-writing and re-trying queries over and over until it found what it wanted, and vector search with a little work up front was worlds better (though it was expensive)

No one runs into machine gun fire to bring democracy to someone else I can assure you of that.

Can’t speak for others but the most damaging part of direct combat is watching people that you consider to be family die or be maimed.

I’ll never understand why LinkedIn is making these product choices. They have a legitimate moat, I and basically every other tech worker is exclusively on LinkedIn. They devalue their product by chasing engagement in service of ads, while allowing things like juicebox to scrape and steal their most valuable asset.

This is short term, make it to next quarter thinking. You couldn’t pay me to use their AI product, they should focus on making that actually work, get the slop out of the feed, and be professional and distinct from TikTok, Facebook, instagram etc.

Flex tier is just as cheap but you don't need to deal with batching or waiting. I see it completing in about ~3x-10x the latency of normal calls, so in seconds. You can always route back to regular if you're getting 429'd on flex too.

Flex is also easier to get caching to work, there is a little futzing around with OAI's implicit caching but if you do the upfront work you can get haiku quality responses with caching in close to realtime for 1/7th the cost and you don't need to design a polling loop to check for batch completions

Haven't worked in AWS but as a customer of AWS, GCP, and Azure there is a quite a difference between running workloads that are doing long-running computations on AWS (on demand is so rock solid that loosing a box is a rarity, I run on spot and still complete almost all of my runs) and running on Azure's "on-demand" style offering feels like spot (or worse) on AWS.

Great news, the fine is so small doesn’t matter, but curing the wrong does. My hope is this standard will apply to modern cars as well, repair manuals and the software tools to interact with the cars are also heavily restricted by the manufacturers.

If you haven’t talked to Anthropic support yet you’re in for a surprise. I’m an engineer at a company with an enterprise contract, Anthropic people in our slack and it took me a month to get a response on my support request, I just decided it wasn’t worth it and bought a second phone number rather than wait.

Performance. A safe Swiss Army knife function that did most of this would be slow because of the internal branching you’d need to be safe, and because there is developer intent in the selection of these functions. I’d rather have the choice and clear dev intent when I see the function used when reading code.